Bits Under ZK-LLM: Evaluating Zero-Knowledge-Friendly Quantization for Verifiable Private LLM Inference
The first systematic study of ZK-friendly quantization for LLMs is presented, formalizing the definition of ZK-friendly quantization and capturing the properties required for ZK proof generation and evaluating nine language models across a broad design space of weight, activation, and nonlinear lookup table precision.
Taeung Yoon, Yu-Peng Zhang, Xiao-Jing Liao
· 0 citations